So, the more the target is underrepresented in the original sample, the more the chance to find right answer decreases. Greater emphasis should be probably given, when planning a clinical trial and when interpreting its results, to the great impact that the molecular heterogeneity of tumors, affecting sensitivity to the experimental treatment, may have on the results of a clinical trial [28]. This concept has been never taken into account in the planning and

the analysis of clinical trials with cytotoxic agents, but it should be necessarily considered Selumetinib purchase in clinical trials with molecularly targeted agents. In a simplified situation, in which the whole population of patients is divided in two distinct genotypes (A and B) – where genotype A is characterized by sensitivity to the experimental treatment producing in this group an outcome better than in the control group, and the genotype B is characterized by absence of difference in efficacy between experimental and standard treatment – the higher the proportion of patients

with genotype B in the study sample, the lower the power of the clinical GDC-0449 chemical structure trial to show a positive result. The statistical power of the study is even lower if we postulate that the genotype B determines a detrimental effect of experimental treatment compared to control. Also in the case that the targeted population is well represented, and the trial gives positive results in favor of the new drug, this means that this effect is driven by that subset of patients, anyway administering the treatment to many patients who do not really benefit. Moreover, Rebamipide the subgroup analysis process itself is biased by many risks of data distortion. According to the brilliant paper

published by Lagakos et al, if you test 10 subgroups, your chance to occur into more than 3, more than 2, and more than 1 false positive results is around 2%, 9% and 40% [29]. Any ‘Post hoc’ exploratory subgroup analyses (i.e. the comparison of experimental and standard treatment separately in subgroups of patients identified by the biomarkers status, without a priori planned hypotheses) is a dangerous procedure, because of the high risk of both false positive and false negative results [30]. Importantly, comparison of treatment and control should not be performed separately in each subgroup, but formal test of interaction should be performed [30]. Of course, results of tests for interactions are likely to be convincing only if they were specified at the start of the study. In any study that presents subgroup analyses it is important to specify when and why the subgroups were chosen [30, 31].